rare‐variant associaon tesng for exome data · kardia_sequence kernel association test author:...

Rare‐VariantAssocia/onTes/ngforExomeData

TheSequenceKernelAssocia/onTest

SequenceKernelAssocia/onTest(SKAT)

•  Gene‐level(orSNPset)analysisapproach•  Testsanassocia/onbetweenSNPsetsandcon/nuousordiscretephenotypes

•  BypassestheproblemofdifferenttaggingSNPsbeingassociatedwithoutcomesofinterestacrossethnicgroups

SKATMainEffectsModel

•  Yi=outcomeforsubjecti•  α0=interceptterm•  Xi=vectorofnon‐gene/ccovariates•  Gi=vectorofgenotypes•  εi=errorterm;followsanydistribu/onwithmean0andvarianceσ2

•  Assumeeachβj,j=1,…,p,followsanarbitrarydistribu/onwithmean0andvariancewjτ– Wheretheweights(wj)arespecifiedbytheuser

3

SKATbasics•  Tes/ng

H0:β=0isequivalenttotes/ngH0:τ=0

•  Thescoretestforvariancecomponentinthecorrespondingmixedmodelisoftheform:

Qρ=(1–ρ)Qs+ρQB

– whereρistheparameteroftheunifiedtest,QSisateststa/s/cofSKAT,andQBisascoreteststa/s/cofweightedburdentest

Kernel

•  Therearepre‐specified6typesofkernels:– "linear“– "linear.weighted"– "IBS“–  "IBS.weighted“– "quadra/c"– "2wayIX"

•  Youcanuseoneofthemoryoucangiveyourownkernelmatrixasaparameter.

DefaultKernel

•  Thedefaultkernelistheweightedlinearkernel

•  Thekernelmatrixfortheweightedlinearkernelis K=GWWG

– WhereGisthenxpmatrixofgenotypedataandWisthepxpdiagonalmatrixoftheweightscorrespondingtoeachvariant.

QSta/s/c

Qρ=(1–ρ)Qs+ρQB

•  TheQsta/s/chasamixtureofchi‐squareddistribu/onunderthenullhypothesesthatcanbeevaluatedexplicitlyandusedasareferencedistribu/ontocomputethep‐values.

Weights

•  ThematrixWisadiagonalmatrixthatcontainstheweightsofthepvariants

•  Goodchoicesofweightscanimprovepower

•  Weightsarepre‐specified

•  Ifweightjislarge,thenthatvariantmakesalargecontribu/ontotheQsta/s/c

•  Upweigh/ngacausalvariant(whichisexpectedtohavealargeeffect)canimprovethepower

•  Wedon’tknowwhichvariantsarecausalandthuswedon’talwaysknowwhichweightstouse

Weights

•  SKATauthorssuggestusing

•  BetaPDF:

WhereBdenotestheBetafunc/on,alphaandbeta(inourweightequa/on,alpha‐1andalpha‐2)areshapeparametersandthefunc/onisevaluatedwhenx=MAFj

Results

•  TheQsta/s/candassociatedp‐valuewilltellusiftheSNPsetisassociatedwiththeoutcome.

•  H0:τ=0,assesseswhetherthereisanyvarianceintheSNPset(Bjs)fromthemeanof0inany(+/‐)direc/on

Example

•  Theassocia/onbetweenSNPsintheChr9p21regionandtheGeneExpression

•  Predictor:SNPsintheChr9p21region(297SNPs)•  Outcome:geneexpressionofthegenesacrossthewholegenome

•  SingleSNPAssocia/onTest– Geneexpression=singleSNP+random(family)(297tests)

•  SequenceKernel(SNPset)associa/on(SKAT)– Geneexpression(anerfamiliaradjustment)=AllSNPs(1test)

SingleSNPAssocia/onAnalysisBetweenCDKN2BASandSNPsintheChr9p21

SingleSNPAssocia/onAnalysisBetweenCDKN2BASandSNPsintheChr9p21

CDKN2BAS(SKATp=0.429)SNP MAF B_SNP_P_CDKN2BAS

rs7865618 0.446317 8.58E‐06

rs634537 0.44901 8.93E‐06

rs2157719 0.495097 9.34E‐06

rs613312 0.439054 1.07E‐05

rs615552 0.453232 1.09E‐05

rs543830 0.438996 1.11E‐05

rs599452 0.438958 1.13E‐05

rs564398 0.440362 1.28E‐05

rs679038 0.439493 1.51E‐05

rs944801 0.459316 1.52E‐05

Specifyingweights

•  Beta(1,25)– Upregulaterarevariantsanddownregulatecommonvariants

•  Beta(1,1)– Equalweightstoallvariants

•  Beta(0.5,0.5)– Madsen & Browning weight

Betadistribu/ons

ApplyingDifferentWeighttoCDKN2BAS

Beta(1,25) Beta(1,1) Beta(0.5,0.5)

SKATpvalue 0.429 0.0020 0.0021

Examplep‐valuesAdjus/ngWeightChangesResultsDrama/cally:

TopResultswithWeightbeta(0.5,0.5)

Transcript NPvalue

beta_1_25Pvalue

_beta_1_1Pvalue_

beta_0.5_0.5

ENST00000301908 801 0.036968216 0.000178972 0.000130954

ENST00000370551 801 0.684927084 0.000178319 0.000217497

ENST00000412318 801 0.112609677 0.000302452 0.00026976

ENST00000497037 801 0.241666486 0.000345966 0.00034696

Conclusion

•  ChoosingappropriateweightisveryimportantinSKAT

•  Beta(1,25)givesverylisleweight,ifany,tothecommonvariants

•  Beta(1,1)hasverylislepowerpickingupsignalsfromrarevariants

•  Beta(0.5,0.5)canpickupsignalsfrombothcommonandrarevariants,butsuffersfromlowerpower.

NextSteps

•  We’vebeenworkingwithShawnLeetotesthisSKATprogramsfor:– Gene‐environmentkernels(currentlyunweighted)– Gene‐genekernels(currentlyunweighted)– Meta‐analysissubrou/nes(Meta‐SKAT)

– Modifica/onsforfamilydata